Method and audio processing system for wind noise suppression

Through the combination of wind noise suppressor module and neural network, wind noise is dynamically processed, and the problems of stroke noise residue and distortion in the existing technology are solved, achieving more effective wind noise suppression and audio content improvement.

CN118922884APending Publication Date: 2024-11-08DOLBY LABORATORIES LICENSING CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202380026675.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2022-12-15
Filing Date
2023-03-08
Publication Date
2024-11-08

Smart Images

  • Figure CN118922884A_ABST
    Figure CN118922884A_ABST
Patent Text Reader

Abstract

The present disclosure relates to a method and system (1) for suppressing wind noise. The method comprises obtaining an input audio signal (100, 101 ') comprising a plurality of consecutive audio signal segments (101, 102, 103, 101', 102 ', 103'), and suppressing wind noise in the input audio signal with a wind noise suppressor module (20) to produce a wind noise reduced audio signal. The method further comprises using a neural network (10) trained to predict a gain set for reducing noise in a given sample of the input audio signal (100, 100 ') of the input audio signal (100, 100') wherein a noise-reduced audio signal is formed by applying the gain set to the input audio signal (100, 101 '), and mixing the wind-noise-reduced audio signal and the noise-reduced audio signal with a mixer (30), to obtain an output audio signal with suppressed wind noise.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Application No. 63 / 432,996 filed on December 15, 2022, U.S. Provisional Application No. 63 / 327,030 filed on April 4, 2022, and International Patent Application No. PCT / CN2022 / 080242 filed on March 10, 2022, all of which are incorporated herein by reference in their entirety. Technical Field

[0003] The invention relates to a method and an audio processing system for wind noise suppression. Background Art

[0004] There are many types of unwanted noise that can be present in audio signals, including non-stationary noise (such as traffic noise or wind noise) and stationary noise (such as white noise or pink noise). Any noise reduces the signal-to-noise ratio of the audio signal and degrades the perceived quality of the audio signal. For example, at high noise levels, the intelligibility of speech content decreases and / or the rendering of spatial audio objects becomes less accurate. Noise caused by wind (i.e., wind noise) is particularly disruptive to many types of audio content, including speech.

[0005] Wind noise typically appears in audio signals recorded by headphones (such as wireless binaural headphones), external microphones, or mobile phones when a user is moving quickly through the air (such as when riding a bicycle) or encounters windy conditions outdoors. Wind noise is unpredictable and may suddenly appear and disappear in the audio content, creating an uncomfortable listening experience for the listener while also masking desired audio content in the audio signal. In general, most of the spectral energy of wind noise is located in the lower audible frequencies below 2kHz, which unfortunately overlaps with a portion of the frequency band associated with human speech, making wind noise particularly disruptive to speech and causing problems for applications such as telephony or teleconferencing.

[0006] Various solutions have been proposed to reduce wind noise. When recording audio signals with professional equipment (e.g. using a dedicated microphone such as a boom microphone), different types of physical filters are installed on the microphone to protect it from strong winds reaching the sensitive recording elements of the microphone. Examples of such physical filters are different windscreen filters, pop-filters or mufflers. However, in many applications, the microphone is small and integrated, for example, in the earpiece of a set of headphones or in a mobile phone, which makes it difficult or impossible to use these usually large and bulky physical filters.

[0007] To this end, different audio processing schemes have been proposed to reduce wind noise in audio signals. In some such processing schemes, a wind detector and a wind suppressor are used to form a noise suppression system that operates on two audio signals. The wind noise detector has a plurality of analyzers, such as a spectrum slope analyzer, a ratio analyzer, a coherence analyzer, a phase variance analyzer, etc., wherein the detection results of each analyzer are weighted together to form a total wind noise detection result for each of the two audio signals. The wind suppressor has a calculation unit and a mixer, the calculation unit calculates a ratio based on the wind noise detection result for each of the two audio signals, and the mixer mixes the two audio signals based on the result of the wind noise detection and the ratio of the calculation unit.

[0008] According to other proposed processing schemes, a trained neural network is used to predict the gain for removing noise from a mono audio signal. When processing two audio signals (e.g., a stereo audio signal pair), each audio signal is processed separately, and the maximum gain predicted for either audio signal is applied to both audio signals to minimize distortion and maintain the perceived position of spatial audio objects. Since this noise reduction technique is aggressive, a remixing module is also used, which reintroduces the audio signal by mixing the original (noisy) audio signal with the noise-reduced audio signal. Summary of the invention

[0009] A disadvantage of existing audio processing solutions for wind noise reduction is that when wind noise is present in only one of the two audio signal outputs, the output audio signal will still contain a high level of residual noise. To circumvent these problems, more aggressive noise processing techniques can be used in conjunction with a remixer, which will reintroduce some of the original audio signal to mitigate acoustic distortion. However, when the wind noise is strong, reintroducing the original audio signal will quickly reintroduce significant levels of wind noise into the audio signal. Therefore, there is a need for an improved method of suppressing wind noise that overcomes at least some of the above-mentioned disadvantages.

[0010] A first aspect of the present invention relates to a method for suppressing wind noise, comprising obtaining an input audio signal comprising a plurality of continuous audio signal segments. The method further comprises suppressing wind noise in the input audio signal with a wind noise suppressor module to produce an audio signal with reduced wind noise, the wind noise suppression module comprising a high pass filter and using a neural network trained to predict a set of gains to reduce noise in a given sample of the input audio signal of the input audio signal, wherein the noise-reduced audio signal is formed by applying the set of gains to the input audio signal. The method further comprises mixing the wind noise-reduced audio signal and the noise-reduced audio signal with a mixer to obtain an output audio signal having suppressed wind noise.

[0011] The wind noise suppressor module can be any wind noise suppressor that performs some filtering or masking on the input audio signal to remove wind noise. Therefore, the resulting wind noise-reduced audio signal is a processed version of the input audio signal with wind noise removed. The wind noise-reduced audio signal may still have one or more other types of noise, such as static white noise and dynamic traffic noise. The neural network is a noise suppression neural network, or a source separation neural network, trained to separate the required audio content (e.g., speech or music) by suppressing all types of noise or unwanted audio content. Therefore, the resulting noise-reduced audio signal is a processed version of the input audio signal, in which one or more types of noise are reduced. It is conceivable that the gain prediction neural network removes static noise as well as and dynamic noise, such as wind noise.

[0012] The inventors have realised that by mixing a wind noise suppressed audio signal with a noise suppressed audio signal, wind noise suppression is achieved without introducing undesirable distortions. Furthermore, the remixing problem is also solved since the original input audio signal is not reintroduced.

[0013] Using only a neural network may introduce unwanted artifacts, as aggressive noise reduction may result in noticeable distortion. In addition, the neural network may remove important ambient audio content that is important for immersion. On the other hand, the wind noise suppressor will suppress wind noise and leave all other types of noise, which are not needed and are important for providing an immersive effect. By combining the two processed audio signals in a mixing ratio, the resulting output audio signal will reduce wind noise due to the processing of the wind noise suppressor and the neural network, both processing schemes leaving the desired audio content, the wind noise suppressor leaving noise not related to the wind, and the neural network also leaving a portion of these noises. Therefore, the resulting output audio signal inherently suppresses wind noise and enhances the desired audio content, while other types of noise content are partially suppressed. Any distortion introduced by the aggressive neural network noise reduction is compensated by the output audio signal of the wind noise suppressor, which is less prone to introducing distortion.

[0014] In some embodiments, the method further comprises determining a wind noise indicator for each segment of the input audio signal with a wind noise detector, the wind noise indicator indicating at least one of a probability and an amplitude of wind noise in each segment. Optionally, the method further comprises determining a wind noise state based on the wind noise indicator. In some embodiments, the wind noise indicator is equal to the wind noise state. In some embodiments, the wind noise state is obtained by smoothing the wind noise metric to avoid rapid fluctuation steering due to rapid fluctuations in the wind noise metric.

[0015] The wind noise indicator or wind noise status can be used to control at least one of the following processes: (A) suppressing wind noise with a wind noise suppressor, (B) the manner in which a set of gains is applied to an input audio signal to form a noise-reduced audio signal, and (C) mixing the noise-reduced audio signal with the wind-noise-reduced audio signal. This means that the wind noise suppression method dynamically adapts to the current wind noise conditions.

[0016] In some embodiments, each of the input audio signal, the wind noise reduced audio signal, and the noise reduced audio signal includes two audio channels, and the method further includes providing the wind noise status to the gain guidance module if the wind noise status of the two channels exceeds a first threshold level, or if the difference between the wind noise conditions of the two channels is below a second threshold level, determining a common gain set based on a predicted gain set for at least one of the two channels, and applying the common gain set to the two channels. Otherwise, the method includes applying each individual gain set to the corresponding channel.

[0017] Therefore, the method avoids applying different gain sets when the difference in wind noise between channels is small or the channels contain similar amounts of wind noise. This ensures that any spatial effects of the two channels are not distorted when the wind noise levels in the two channels are similar or low. On the other hand, when one of the channels includes strong wind noise, or when there is a large difference in the amount of noise between the two channels, the method applies different gain sets to different channels to suppress wind noise when it is most needed, at the expense of possible spatial distortion.

[0018] In some embodiments, determining the wind noise state includes providing a wind noise indicator to a state machine having at least two states, a no wind noise state NWN and a wind noise hold state WNH. The state machine transitions to the WNH state in response to detecting a first number of subsequent segments associated with the wind noise indicator exceeding a high threshold, and outputs a high wind noise state at least until the next state change, and the state machine transitions to the NWN state in response to detecting a second number of subsequent segments associated with the wind noise indicator below a first low threshold, and outputs a low wind noise state at least until the next state change. Optionally, the state machine includes four states.

[0019] Through the state machine, a stable wind noise state is obtained, which ensures that the control of system components based on the wind noise state is not affected by oversteering, which may cause noticeable acoustic artifacts due to fast handling changes.

[0020] In some embodiments, the input audio signal, the wind noise reduced audio signal, and the noise reduced audio signal include two audio channels, and the method further includes determining a differential metric of the two audio channels with a dual mono detector, the differential metric indicating a difference in spectral energy in at least one frequency band of the two audio channels. If the differential metric is less than a differential threshold, the method includes determining a wind noise indicator based only on the mono characteristics of each individual audio channel. Otherwise, the method includes determining a wind noise indicator based on both the mono characteristics of each audio channel and the differential characteristics associated with the two audio channels.

[0021] Examples of monophonic features determined for a single channel are spectral slope of one or more frequency bands and power density centroid of one or more frequency bands.Examples of differential features are differential measures of spectral power, coherence and phase of one or more corresponding frequency bands of two audio channels.

[0022] Therefore, by determining the degree of dissimilarity of the channels, and adjusting the method used to determine the wind noise indicator accordingly, a robust and accurate wind noise indicator can be extracted regardless of the level of dissimilarity between the channels of the input audio signal. For example, without distinguishing between highly similar and highly dissimilar audio channels, the differential features may result in false positives indicating strong wind noise even if the channels contain very low or no wind noise, since the channels contain very similar content. Furthermore, since only monophonic features are determined for very similar audio signals, computational complexity is reduced.

[0023] According to a second aspect of the present invention, there is provided a wind noise suppression system, comprising a processor and a memory coupled to the processor, wherein the processor is adapted to execute the method of the first aspect of the present invention.

[0024] According to a third aspect of the present invention, there is provided a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to perform the method according to the first aspect of the present invention.

[0025] According to a fourth aspect of the present invention, there is provided a computer readable storage medium storing a computer program according to the third aspect of the present invention.

[0026] The invention according to the second, third and fourth aspects has the same or equivalent advantages as the invention according to the first aspect.Any function described in relation to the wind noise suppression method may have a corresponding feature in the wind noise suppression system, and vice versa. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The invention will now be described in more detail with reference to the accompanying drawings, which show currently preferred embodiments of the invention.

[0028] Figure 1A system for suppressing wind noise according to some embodiments is depicted.

[0029] FIG. 2 a schematically illustrates an input audio signal having a single channel in some embodiments.

[0030] Fig. 2b schematically depicts some embodiments of different input audio signals with two channels.

[0031] Figure 3 Depicted is a system for suppressing wind noise using a smoothing module in addition to a dual-mono detector, a wind noise detector, and a smoother, according to some embodiments.

[0032] Figure 4 is a flow chart illustrating a method of suppressing wind noise according to some embodiments.

[0033] Figure 5 A state machine with four states used in a smoothing module according to some embodiments is schematically depicted.

[0034] Figure 6 A system for suppressing wind noise having a context analysis module for enhancing non-real-time processing is depicted in accordance with some implementations. DETAILED DESCRIPTION

[0035] The systems and methods disclosed in this application can be implemented as software, firmware, hardware, or a combination thereof. In a hardware implementation, the division of tasks does not necessarily correspond to division into physical units; on the contrary, a physical component can have multiple functions, and a task can be performed collaboratively by multiple physical components. Computer hardware can be, for example, a server computer, a client computer, a personal computer (PC), a tablet computer, a set-top box (STB), a personal digital assistant (PDA), a cellular phone, a smart phone, a network device, a network router, a switch or bridge, a wearable device (XR or AR or VR or MR headset, audio headset, etc.) or any machine capable of executing instructions (sequentially or otherwise) specifying the actions to be taken by the computer hardware. In addition, the present disclosure will relate to any collection of computer hardware that executes instructions individually or jointly to perform any one or more of the concepts discussed herein.

[0036] Some or all of the components may be implemented by one or more processors that accept a computer-readable (also referred to as machine-readable) code containing a set of instructions that, when executed by one or more processors, perform at least one of the methods described herein. Any processor that is capable of executing a set of instructions (sequentially or otherwise) that specifies an action to be taken is included. Thus, an example is a typical processing system (i.e., computer hardware) that includes one or more processors. Each processor may include one or more of a CPU, a graphics processing unit, and a programmable DSP unit. The processing system may also include a memory subsystem that includes a hard drive, an SSD, RAM, and / or ROM. A bus subsystem may be included for communication between components. During execution of software by the computer system, the software may reside in the memory subsystem and / or within the processor.

[0037] One or more processors may operate as standalone devices or may be connected (e.g., networked) to (one or more) other processors. Such a network may be built on a variety of different network protocols, such as the Internet, a wide area network (WAN), a local area network (LAN), or any combination thereof.

[0038] The software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transient medium). As is well known to those skilled in the art, the term "computer storage medium" includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules or other data. Computer storage media include, but are not limited to, various forms of physical (non-transitory) storage media, such as EEPROM, flash memory or other storage technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media (transitory) typically embodies computer-readable instructions, data structures, program modules or other data in a modulated data signal (such as a carrier wave or other transmission mechanism), and includes any information transfer medium.

[0039] Figure 1A system 1 for suppressing wind noise according to some embodiments is depicted. Components of the system 1 may be computer implemented, including a processor and a memory coupled to the processor. An input audio signal is received and provided to a wind noise suppressor module 20 and a neural network 10, which is trained to predict a set of gains to reduce noise for a given sample of the input audio signal. The predicted set of gains is provided to a gain applicator 11, which applies the gains to the input audio signal and provides the resulting noise-reduced audio signal Y to a mixer 30. The mixer 30 also receives a wind-noise-reduced audio signal X from the wind noise suppressor module 20, and mixes the two audio signals in a mixing ratio specified by a mixing coefficient μ to generate an output audio signal Z.

[0040] The input audio signal includes one or more audio channels. In some embodiments, the input audio signal includes wind noise mixed with the desired audio content, and optionally other forms of noise, such as static white noise, static pink noise and / or non-static traffic noise. The neural network 10 is any type of neural network that is trained to predict a gain set to suppress all types of noise except a target source (e.g., speech or music). The neural network can be implemented as an RNN, a CNN, etc.

[0041] The predicted gain set includes a plurality of gains, each associated with a single frequency band of a segment. The term "gain" is used herein, and it will be understood that "gain" can mean actual gain (i.e., amplification) or attenuation (i.e., reduction in amplitude). Most denoising neural networks 10 are trained to predict gains for all audio content that is different from the desired audio content (e.g., speech or music). For example, if a frequency band in a segment of the input audio signal includes noise, and there is little or no desired audio content, the neural network 10 will output a gain that suppresses the frequency band of the segment. It will be understood that the process is dynamic, and as the spectral distribution of the desired audio content and the noise change over time, the suppressed frequency bands and the degree to which they are suppressed vary from segment to segment.

[0042] The wind noise suppressor 20 is configured to suppress wind noise in the input audio signal. Since wind noise usually occurs in the low frequency range (below 2kHz), the wind noise suppressor includes a high pass filter with a cutoff frequency and a selectable passband variable gain. In some embodiments, as described below, the cutoff frequency and the passband gain are dynamically adjustable (based on the wind noise indicator / state).

[0043] The wind noise suppressor 20 is configured to remove wind noise from the audio signal. Since wind noise is typically present at low frequencies, the wind noise suppressor will typically not remove high-frequency noise, such as the high-frequency components of static white noise or pink noise. On the other hand, the neural network 10 typically performs more aggressive noise suppression, in which all noise content that is not identified as desired audio content is suppressed. This may mean that the neural network 10 generates a gain for suppressing static noise (e.g., white noise) as well as dynamic noise (such as wind noise or traffic noise). Although the neural network 10 can be very effective in isolating the desired audio content, it may cause unnecessary distortion or misclassify the desired audio content as noise, while the analytical wind noise suppressor 20 operates in a more controlled and predictable manner.

[0044] FIG. 2 a schematically depicts an input audio signal 100. The input audio signal comprises a plurality of consecutive segments 101, 102, and 103, each of which comprises a portion of the audio signal 100. The plurality of segments 101, 102, and 103 may form a complete audio file of any duration (e.g., ranging from several minutes if the audio file is a music track to several hours if the audio file is the voice content of a phone call or the soundtrack of a movie). Each segment may have any suitable length. In some embodiments, the length of each segment 101, 102, and 103 is between 2 milliseconds and 50 milliseconds, such as 5 or 10 milliseconds.

[0045] The input audio signal 100 may be represented in the time domain or the frequency domain. The input audio signal 100 may be represented by a time-frequency block or by a filter bank (e.g., a QMF filter bank) known to those skilled in the art. The input audio signal may include one channel or two or more channels.

[0046] In Fig. 2b, an input audio signal 100' comprising two audio channels is depicted. It can be seen that the two audio channels are divided into corresponding segments 101, 102, 103, 101', 102' and 103'. Figure 1 In the wind noise suppression system of FIG. 1 , the two audio channels are processed separately or together. For example, the wind noise suppressor can determine different high-pass filters and apply them to the two channels or a common filter. Similarly, as described below, the neural network can predict a separate gain set for each channel, whereby the gain applicator applies the separate gain set to the corresponding channel or combines the separate gain sets into a common gain set.

[0047] The two audio channels may be any type of audio channels, such as stereo, binaural audio channels or any selection of two arbitrary channels. The stereo audio channels may be, for example, left and right audio channels. It is also envisaged that the stereo audio channels have different stereo presentations, for example formed by a center and a side audio channel. The two audio channels may contain the same or at least very similar audio content (such as in the case of center-panned stereo music content), or audio content containing no or almost no common audio content (such as audio content for a center speaker and a left rear speaker in a 5.1 presentation of a sound file associated with a movie).

[0048] Figure 3 Depicted Figure 1 A system 1 for wind noise suppression, wherein a dual mono detector 40 and a wind noise detector 50 have been added.

[0049] The wind noise detector 50 analyzes the input audio signal and determines a wind noise indicator for each segment of the input audio signal. The wind noise indicator is a measure (e.g., one or more numerical values) indicating whether wind noise is present in the input audio signal. To this end, the wind noise indicator can be a measure of the wind noise amplitude or probability. The wind noise indicator can be a binary value (i.e., indicating the presence or absence of wind noise) or a scalar value (soft score) indicating the amplitude / probability of wind noise in each segment. For example, the scalar value (soft score) can range from 0 to 1, with a smaller value representing a lower wind noise amplitude / probability and a higher value representing a higher wind noise amplitude / probability.

[0050] The wind noise detector 50 extracts several features from the input audio signal and determines a wind noise indicator based on these features. The wind noise indicator can determine a mono feature of each individual channel of the input audio signal and / or a differential feature based on at least two channels in the input audio signal. The wind noise detector 50 can determine at least one of the following features: a spectral slope in one or more frequency bands of a channel, a power centroid in a subband or multiple frequency bands of a channel, a power ratio between two channels in at least one frequency band, a coherence between two channels in at least one frequency band, a phase difference between two channels in at least one frequency band, and determine the wind noise indicator based on segmentation of the input audio signal and the determined (one or more) features of each channel.

[0051] The wind noise detector 50 may also be implemented by conventional machine learning algorithms (such as support vector machines, AdaBoost, or deep neural networks) depending on available computing resources.

[0052] At least one of the wind noise indicator and a wind noise status (which is derived from the wind noise indicator) is provided to at least one of the mixer 30, the wind noise suppressor 20 and the gain applicator 11 to dynamically control the operation of at least one component.

[0053] exist Figure 3 In the embodiment of the present invention, as described in further detail below, the wind noise indicator is provided to the smoother 60, which extracts the wind noise state by smoothing the wind noise indicator, whereby the wind noise state is provided to at least one of the mixer 30, the wind noise suppressor 20, and the gain applicator 11 to dynamically control the operation of at least one component. In some embodiments, the wind noise indicator is alternatively provided to at least one of the mixer 30, the wind noise suppressor 20, and the gain applicator 11 (e.g., when the smoother 60 is not present) to dynamically control the operation of at least one component. It is also contemplated that both the wind noise indicator and the wind noise state can be provided to at least one of the mixer 30, the wind noise suppressor 20, and the gain applicator 11.

[0054] The mixer 30 may receive the wind noise indicator and / or the wind noise status and change the mixing coefficient μ (mixing ratio) accordingly. Figure 3 In FIG. 1 , the wind noise-reduced audio signal from the wind noise suppressor 20 is labeled as X, and the noise-reduced audio signal from the gain applicator 11 is labeled as Y. The mixer 30 mixes the outputs X and Y into the output audio signal Z according to the following formula based on the mixing coefficient μ:

[0055] Z=μY+(1-μ)X (1)

[0056] In other words, the mixing coefficient μ determines the mixing ratio of X and Y.

[0057] The wind noise indicator and / or wind noise status will affect the value of the mixing coefficient μ, thereby emphasizing X or Y in the output Z more or less. In some embodiments, μ switches between two values, such as μ=0.9 and μ=0.6, where the higher value is assigned when the wind noise indicator and / or wind noise status indicates a high level of wind noise, and the lower value is assigned when the wind noise indicator and / or wind noise status indicates a low level of wind noise. It is also envisioned that for a scalar-valued wind noise indicator and / or wind noise status, the mixing coefficient μ is interpolated between two values ​​based on the scalar value.

[0058] Alternatively, the mixing coefficient μ may be static or user specified, where if clearer audio is desired, μ is set to 0.75, and if more ambient sound is desired, μ is set to 0.5. The value of μ may also be selected based on the ambient context.

[0059] The wind noise suppressor 20 may receive a wind noise indicator and / or a wind noise state and perform wind noise suppression based on the wind noise indicator. For example, the properties of the high pass filter may be adjusted based on the wind noise indicator and / or the wind noise state. For a scalar value (soft score) of the wind noise indicator, the high pass filter gain HPF for a given frequency band b may be determined as:

[0060]

[0061] Where A is the maximum attenuation in decibels (for example, a value of -30dB can be used), and the entity STEP is calculated by solving the equation A+STEP*b max = 0, which means STEP = -A / b max , b max is the highest frequency band in which wind noise is detected (e.g., 1.6 kHz). The variable w is a scalar wind noise indicator and / or wind noise state. Therefore, from the high pass filter defined by Equation 2 above, it is clear that below b max The frequency band above b will be suppressed, while max The frequency will not be affected.

[0062] If only a binary wind noise indicator and / or wind noise status is available, the wind noise power spectral density N(N,f) is determined by recursive smoothing of the squared wind noise spectrum magnitude as follows:

[0063] Φ(n,f)=a(n)Φ(n-1,f)+(1-a(n))|N(n,f)| 2 ,f <F (3)

[0064] Where a(n) and F are assigned values, such as a(n) = 0.8 and F = 1.6kHz. Wind noise power |N(n,f)| 2 The wind noise is determined only for wind noise segments associated with high binary wind noise indicators and / or wind noise states. After determining the wind noise power spectral density, the wind noise suppressor 20 can use a classical noise reduction algorithm (such as spectral subtraction) to reduce the wind noise. In other words, applying a high pass filter may include performing spectral subtraction on the estimated wind noise power spectral density.

[0065] The gain applicator 11 can receive a wind noise indicator and / or a wind noise state from the wind noise detector 50, and modify the gain application based on the wind noise indicator and / or the wind noise state. This is particularly beneficial when the input audio signal includes two audio channels, wherein the wind noise detector 50 has determined the wind noise indicator and / or the wind noise state for each segment of each channel, and the neural network has predicted a separate gain set for each channel. The gain applicator 11 will analyze the wind noise indicator and / or the wind noise state of each channel, and determine whether to apply each gain set to the corresponding channel (mode A) or determine a common gain set to be applied to the two channels based on the wind noise indicator and / or the wind noise state from each channel, which is based on the separate gain set of the corresponding channel (mode B). For example, the common gain set can be the maximum or average value of the elements of the two gain sets.

[0066] If the gain applicator 11 determines that the wind noise indicator and / or wind noise status indicates that only one channel has wind noise (exceeding a first threshold level γ1), or if the difference in wind noise indicators and / or wind noise status between channels exceeds a second threshold level γ2, the gain applicator 11 will operate in mode A.

[0067] If the gain applicator 11 determines that the wind noise indicator and / or wind noise status indicates that both channels have wind noise (exceeding the first threshold level γ1), or if the difference in wind noise indicators and / or wind noise status between the channels is below a second threshold level γ2, the gain applicator 11 will operate in mode B.

[0068] For example, the second threshold level γ2 is selected between 0.1 and 0.3, such as γ=0.2. There are also many suitable values ​​for the first threshold level γ1. For example, γ1 can be selected from the interval of 0.1 to 0.6, such as 0.5 or 0.25. However, these values ​​of γ2 and γ1 are merely exemplary, and other values ​​of γ2 and γ1 can be envisioned depending on the properties (e.g., sensitivity) of the wind noise detector 50.

[0069] Mode A ensures adequate wind noise attenuation for channels containing very strong wind noise, while mode B ensures that small amounts of wind noise are not introduced without introducing different filtering of the two channels, which could change the spatial balance of the two channels in an unnecessary and distracting way.

[0070] As described above, the wind noise detector 50 determines the wind noise indicator based on a mono feature and / or based on a differential feature between two channels. If the input audio signal is a mono audio signal, only the mono feature can be used. However, if the input audio signal includes two or more channels, the wind noise detector can use either or both of the mono feature and the differential feature.

[0071] In order to improve accuracy when determining the wind noise index of two input channels, it is important to detect the similarity of the two channels of the input audio signal in terms of audio content. If, according to the differential metric, the channels contain similar or equal audio content, the wind noise detector 50 will only use mono features when determining the wind noise indicator. Such audio channels associated with a low differential metric are referred to herein as dual-mono audio channel pairs, because highly similar channels may be similar to the duality of the same mono audio signal. On the other hand, if the channels are not too similar (i.e., associated with a high differential metric), the wind noise detector 50 will use both mono features and channel differential features for each audio channel.

[0072] It will be appreciated that a pair of audio channels exhibiting a low difference metric will have a lot of common audio content between the channels, while a pair of audio channels having a higher difference metric will have little or no common audio content.

[0073] To this end, the system 1 includes a dual mono detector 40, which is configured to determine whether the two channels of the input audio signal are sufficiently different from each other so that the wind noise detector 50 can use both the channel difference characteristics and the mono characteristics when determining the wind noise indicator. The dual mono detector 40 analyzes the audio channels and determines a difference metric. The difference metric can, for example, indicate at least one of the differences in spectral energy in one or more frequency bands.

[0074] In one embodiment, the dual mono detector 40 operates in the frequency domain (eg, in the Short Time Fourier Transform STFT domain) and calculates the sum S of the absolute values ​​of the spectral differences between the first and second channels as follows:

[0075]

[0076] Where b is a frequency band index ranging from a low frequency band index b1 to a high frequency band index b2, CH1 and CH2 represent the first and second channels, respectively. The dual mono detector 40 also calculates the total energy E1, E2 of each channel separately as follows:

[0077]

[0078] Based on the sum S and the determination of the individual channel energies E1, E2, the dual mono detector 40 determines the normalized sum (ratio) S N as follows:

[0079]

[0080] If the dual mono detector 40 determines that S NIf the total frame energy E1+E2 is less than a first predefined threshold α, and the total frame energy E1+E2 is higher than a second predefined threshold β, then the segment contains sufficiently similar audio channels and can be marked as dual mono. If these criteria are not met, it is determined that the segment contains non-dual mono audio channels. The dual mono detector 40 transmits information to the wind noise detector 50, indicating whether the dual mono detector 40 has determined that the segment is a dual mono segment, which allows the wind noise detector 50 to selectively use only mono features (for dual mono channels) or use both mono features and differential features (for non-dual mono channels).

[0081] Sum S, S N The relationship between the channel energies E1, E2 are examples of differential metrics that can be used to determine whether the channels are similar enough to use only monophonic features.

[0082] In some embodiments, the dual mono detector 40 includes a dual mono counter that counts the number of segments encountered that have sufficiently similar audio content to be marked as dual mono segments. Once the counter reaches a predetermined number N decorr , the entire input audio signal is classified as a dual-mono audio signal, and all subsequent segments are considered dual-mono audio segments. To this end, once the counter reaches N decorr , the dual mono detector 40 may be deactivated, at least until the next audio signal is input to the system 1. The counter of the dual mono detector 40 may be reset each time a segment is detected that does not contain sufficiently similar channels to be classified as a dual mono segment.

[0083] For example, the low frequency band index b1=0, the high frequency band index b2=512, and the first predefined threshold α=10 -4 , the second predefined threshold β=10 -8 , and N decorr =40.

[0084] Optionally, the system 1 for wind noise suppression cooperates with a smoothing module 60, which smoothes the wind noise indicator. Based on the smoothed wind noise indicator, a wind noise state is determined, which state is used to control components of the system 1. In other words, the wind noise state is a processed version of the wind noise indicator, wherein either or both of the wind noise indicator and the wind noise state can be used as control components as indicated above. The wind noise state can, for example, be equal to a smoothed version of the wind noise indicator. By further reference to Figure 4 Flowchart in, a method for suppressing wind noise will now also be described in detail.

[0085] In step S1, the input audio signal is obtained and provided to the wind noise suppressor 20, the neural network 10 and the gain applicator 11. In addition, the input audio signal is provided to the dual mono detector 40, which determines in step S2a (for example, according to the above equations 4-7) whether the input audio signal includes similar (dual mono) audio channels. If the audio channels are similar, they are dual mono, and the method proceeds to S2b and determines a wind noise indicator for each channel based only on the mono feature. On the other hand, if it is determined that the audio channels are different, that is, not dual mono, the method proceeds to S2c and determines a wind noise indicator for each audio channel based on the mono feature and the channel differential feature.

[0086] Optionally, the wind noise indicator is provided to the smoothing module 60 , and the smoothing module 60 smoothes the wind noise indicator in step S3 to obtain a more stable wind noise state.

[0087] The smoothing module 60 smoothes the wind noise indicator over a plurality of adjacent segments to obtain a more stable wind noise state from the wind noise indicator. The wind noise state then replaces the wind noise indicator and is provided to at least one of the mixer 30, the wind noise suppressor 20, and the gain applicator 11 to enable the wind noise indicator to be used in a manner similar to that described above in conjunction with the wind noise indicator. Figure 3 These components are dynamically controlled in a manner guided by emphasized wind noise indicators.

[0088] If the wind noise indicator is a scalar value for each segment, the smoothing performed by the smoothing module 60 may require averaging the wind noise indicator over multiple adjacent segments. In addition, if the wind noise indicator is a binary wind noise indicator, smoothing may also be performed. For example, a high binary level is represented by 1 and a low binary level is represented by 0, whereby smoothing across multiple frames allows the wind noise state to take a fractional value between 0 and 1. Although smoothing using an average across adjacent segments will eliminate the most rapid changes in the guidance of different components in the wind noise suppression system, there are still situations where even an average across many segments may result in rapid switching between, for example, applying a common gain set or applying a single gain set. To this end, as described in more detail below, the smoothing module may employ a state machine.

[0089] In step S4, the input audio signal is provided to the wind noise suppressor 20, which suppresses the wind noise and generates an audio signal X with reduced wind noise. The wind noise suppressor 40 can be controlled by the wind noise indicator to apply a high pass filter, as described in equation 2, which changes dynamically with the wind noise indicator or the wind noise state. The wind noise suppressor 40 can determine and apply a separate high pass filter for each channel, or determine and apply a common high pass filter for two channels. For example, a common high pass filter is an average filter gain for each frequency band.

[0090] At step S5a, the input audio signal is provided to the neural network 10, which predicts a set of gains for each channel, wherein at step S5b, the gain set is applied by the wind noise applicator 11 to reduce the noise in the input audio signal, thereby generating a noise-reduced audio signal Y. As described above, the gain applicator can be controlled by a wind noise indicator or wind noise status to determine whether a gain set should be applied to a channel individually, or to determine a common gain set to be applied to both channels.

[0091] At step S6, the mixer 30 combines the wind noise reduced audio signal X and the noise reduced audio signal Y at a mixing ratio. The mixing ratio is determined by the mixing coefficient μ and can be dynamically adjusted using a wind noise indicator or wind noise status as described in Equation 1 above.

[0092] The smoothing module 60 may include Figure 5 The state machine 61 shown is used to smooth the wind noise indicator to a more stable wind noise state. The state machine includes four states: a no wind noise NWN state, a wind noise attack (attack) WNA state, a wind noise hold WNH state, and a wind noise release WNR state. In each state, the state machine outputs a high (e.g., 1) or a low (e.g., 0) wind noise state, and the state machine will continue to output the same wind noise state until the state changes. Although some states, such as the NWN state and the WNH state, are always associated with the same wind noise state (a low state for NWN and a high state for WNH), some states (such as the WNA state) may be associated with a high or low wind noise state depending on which state the state machine is transitioned from.

[0093] The state machine 61 is evaluated once for each segment (ie once for each updated value of the wind noise indicator), and for embodiments with multiple channels, one state machine 61 is employed to smooth the wind noise indicator for each channel.

[0094] The state machine 61 is configured to start from an initial state of NWN or WNH.

[0095] Without loss of generality, Figure 5 The state machine 61 starts by switching to the NWN state through L1 and outputs the low wind noise state. For each new segment and each associated wind noise indicator, the state machine 61 will check whether the wind noise indicator is greater than the high threshold T high If the wind noise indicator remains below the high threshold T high , the NWN state will be maintained and the state machine continues to output the low wind noise state.

[0096] If the state machine 61 detects that T is exceeded while in the NWN state highIf the wind noise indicator is , the state machine 61 switches to the WNA state through L2 and will continue to output the low wind noise state. When the state machine 61 enters the WNA state, it also starts the attack counter 501, which detects the wind noise indicator exceeding the high threshold T. high As long as the attack counter 501 is below the first predetermined number N acc , the WNA state will be maintained. If the wind noise indicator is lower than the first low threshold T low1 , then the attack counter 501 is reset to zero.

[0097] When the attack counter 501 reaches N acc When the state machine 61 switches to the WNH state through H1, and the output wind noise state changes from the low wind noise state to the high wind noise state. In the WNH state, the state machine 61 checks whether the wind noise indicator is higher than the second low threshold value T low2 , and as long as the wind noise indicator is above the second low threshold T low2 , the state machine 61 remains in the WNH state and outputs a high wind noise state. low2 Greater than the first low threshold T low1 .

[0098] If the state machine 61 detects that the wind noise indicator is below the second low threshold value T while in the WNH state low2 , the state machine 61 switches to the WNR state through H2, while continuing to output the high wind noise state. At the same time, entering the WNR state triggers the release counter 502, which releases the counter 502 for the number of times the signal is lower than the second low threshold T low2 As long as the release counter 502 is lower than the second predetermined number N low , the counter continues to operate and the state machine 61 will remain in the WNR state and output the low wind noise state.

[0099] Once the release counter 502 reaches the second predetermined number N low , the state machine 61 switches to the NWN state through L3 and starts to output the low wind noise state.

[0100] If the state machine 61 in the WNR state detects that the wind noise indicator is greater than the high threshold, the state machine 61 transitions to the WNA state via H3 while maintaining the output as the high wind noise state. This is in contrast to when the state machine 61 enters the WNA state from the NWN state along L2, where the output is fixed to the low wind noise indicator. In some embodiments, the first predetermined number N acc Higher than the second predetermined number N low , which causes the state machine 61 to be biased toward the NWN state. low =20, Nacc =40.

[0101] In some embodiments, the second lower threshold T low2 is the average of the first low threshold and the second low threshold. For example, T low1 =0.04, T high =0.2, T low2 =0.12.

[0102] In addition, in some embodiments, it is beneficial to use more than two smoothing techniques. For example, the wind noise indicator is first smoothed by averaging adjacent segments, running a smoothing window on the wind noise indicator, or determining a historical weighted sum (where the current wind noise indicator is given the maximum weight, earlier wind noise indicators are given less weight, and the earliest wind noise indicator has the smallest weight). This traditional smoothing process generates a smoothed wind noise indicator. The smoothed wind noise indicator is then provided to the state machine 61, which determines a more stable wind noise state based on the smoothed wind noise indicator.

[0103] The above-described system 1 for wind noise suppression is applicable to both real-time (causal) processing and non-real-time (non-causal) processing implementations. However, in some implementations, a context analysis module 70 is added to the wind noise suppression system to further enhance non-real-time processing.

[0104] The context analysis module 70 aggregates the segment-by-segment outputs of the wind noise detector 50 (i.e., the wind noise indicator) and the smoother 60 (i.e., the wind noise state extracted from the wind noise indicator, e.g., by a state machine) over many segments (e.g., all segments in an audio file) and determines a global wind noise metric W for all segments. p The context analysis module 70 may, for example, determine a weighted average of wind noise metrics and / or wind noise conditions, wherein the global wind noise metric is based on (or equal to) the weighted average. p is a scalar value between 0 and 1, where lower values ​​represent lower wind noise magnitude / confidence for the segment being analyzed and higher values ​​represent higher wind noise magnitude or confidence for the segment being analyzed.

[0105] In some embodiments, a wind noise suppression system 1 ' is used that is configured to utilize auxiliary sensors and / or auxiliary classifier units. The auxiliary sensors may be, for example, acoustic sensors (e.g., microphones), environmental sensors, GPS receivers, motion sensors, vibration sensors, or any other type of sensor typically available on user devices such as smartphones, earbuds, smart watches, tablets, which are examples of devices that may implement the wind noise suppression method. The auxiliary classifier unit may be, for example, one of a music classifier, a speech classifier, and an acoustic scene classifier. The classifier unit may be implemented in software and / or hardware and is typically available on the above-mentioned example devices.

[0106] exist Figure 6 In FIG. 7 , this is shown as the context analysis module 70 accepting auxiliary data, where the auxiliary data is the output of the auxiliary sensors and / or auxiliary classifier units described above. For example, the auxiliary data originates from a voice activity detection, VAD, classifier unit, music classifier unit, and / or acoustic scene classifier unit. The context analysis module 70 utilizes the auxiliary data to determine a more accurate global wind noise metric W p For example, if the auxiliary data indicates that the segment may include speech and / or music (by receiving a high VAD value or a high music classifier confidence in the auxiliary data) and includes wind noise (due to a high wind noise indicator), the context analysis module 70 calculates W p If the segment does not include speech and / or music (by receiving a low VAD value and / or a low music confidence in the auxiliary data), but does include wind noise (due to a high wind noise indicator), the context analysis module 70 may give the segment a lower weight when calculating W because it is unclear whether the segment is music, speech, or wind. On the other hand, if the segment does not include speech and / or music (by receiving a low VAD value and / or a low music confidence in the auxiliary data), but does include wind noise (due to a high wind noise indicator), then the context analysis module 70 may give the segment a lower weight when calculating W. p A higher weight can be given to the segment because there is a higher confidence that the segment includes wind noise.

[0107] Then, the global wind noise metric W p At least one of the wind noise suppressor 20' and the mixer 30' is provided with a context modification, which employs a modified type of processing compared to the above-described processing.

[0108] The context-modified wind noise suppressor 20 ′ receives the global wind noise metric W p , and the high-pass filter is determined according to the following equation:

[0109]

[0110] The difference compared to the high-pass filter described in Equation 2 is that the gain (attenuation) is expressed in terms of the global noise metric indicator W pFor example, if the context analysis module 70 determines that the analyzed segment is generally free of wind noise, then the global wind noise indicator W p will be close to zero, and there will be little or no attenuation in the low frequency stopband. On the other hand, if the context analysis module 70 determines that the analyzed segment contains wind noise as a whole, the global wind noise indicator W p will be close to 1, and there will be a high level of attenuation in the low frequency stop band.

[0111] The context modified mixer 30 ′ receives the global wind noise indicator W p , and based on W according to the following formula p Determine the mixing coefficient μ p :

[0112] μ p = W p *μ1+ (1-W p )*μ0 (9)

[0113] Wherein μ1 is higher than μ0. Usually, μ1 is between 0.6 and 0.9, such as 0.75, and μ0 is between 0.4 and 0.6, such as 0.5.

[0114] The mixing performed by mixer 30 is then given by the following equation:

[0115] Z= μ p Y+ (1-μ p )X (10)

[0116] Where Z is the output audio signal, X is the wind noise suppressed output audio signal from the context modified wind noise suppressor 20', and Y is the audio signal output by the gain applicator module 11 which has applied the set(s) of gains predicted by the neural network 10. Thus, for the plurality of segments forming the audio file, the mixing coefficients μ p remains a constant for the interpolation between μ1 and μ0. It will be appreciated that the processing performed using the context analysis module 70 is no longer causal, as later (later) segments will affect W used to process the current (earlier) segment. p .

[0117] Unless otherwise specifically stated, as will be apparent from the following discussion, throughout the disclosure discussion, terms such as "processing," "computing," "calculating," "determining," "analyzing," etc., are used to refer to the actions and / or processes of computer hardware or computing systems or similar electronic computing devices that manipulate and / or transform data represented as physical (e.g., electronic) quantities into other data similarly represented as physical quantities.

[0118] It should be understood that in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes combined in a single embodiment, figure, or description thereof to simplify the present disclosure and help understand one or more of the various creative aspects. However, this method of disclosure should not be interpreted as reflecting the intention that the claimed invention requires more features than those explicitly stated in each claim. On the contrary, as reflected in the following claims, the creative aspects lie in less than all the features of a single aforementioned disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into this detailed description, and each claim exists independently as a separate embodiment of the present invention. In addition, although some embodiments described herein include some features included in other embodiments but do not include other features, the feature combinations of different embodiments are intended to be within the scope of the present invention and form different embodiments, as understood by those skilled in the art. For example, in the following claims, any claimed embodiment can be used in any combination.

[0119] In addition, some embodiments are described herein as a combination of methods or method elements that can be implemented by a processor of a computer system or by other devices that perform functions. Therefore, a processor with instructions for performing such methods or method elements forms a device for performing methods or method elements. Please note that when the method includes several elements, such as several steps, the order of these elements is not implied unless otherwise specified. In addition, the elements of the device embodiments described herein are examples of devices for performing the functions performed by the elements to implement embodiments of the present invention. In the description provided herein, many specific details are set forth. However, it will be understood that embodiments of the present invention can be practiced without these specific details. In other cases, in order not to obscure the understanding of this description, well-known methods, structures and techniques are not shown in detail.

[0120] Those skilled in the art recognize that the present invention is by no means limited to the preferred embodiments described above. On the contrary, within the scope of the appended claims, many modifications and variations are possible. For example, any dynamic control based on the wind noise state (as a smoothed version of the wind noise indicator) can be replaced by the wind noise indicator, regardless of whether the wind noise indicator is a scalar value or a binary metric. Although a smoother and more stable wind noise state over time has a beneficial effect in not causing rapid and significant changes in audio processing, it will be understood that for some audio signals, the use of a wind noise indicator directly provides sufficient performance. In addition, although most of the exemplary embodiments are described using audio signals of one or two channels, the method and system can also be applied to three or more channels in a similar manner. For example, the processing can be based on paired selections between multiple channels.

Claims

1. A method for suppressing wind noise, comprising: Acquiring (S1) an input audio signal (100, 100') comprising a plurality of consecutive audio signal segments (101, 102, 103, 101', 102', 103'); suppressing (S4) wind noise in an input audio signal (100, 100') with a wind noise suppressor (20) module to produce an audio signal with reduced wind noise, the wind noise suppressor module comprising a high pass filter; using (S5a) a neural network (10) trained to predict a set of gains for reducing noise in a given sample of an input audio signal (100, 100') of the input audio signal (100, 100'), wherein a noise-reduced audio signal is formed by applying the set of gains to the input audio signal (100, 100'); as well as The wind noise reduced audio signal and the noise reduction audio signal are mixed (S6) with a mixer (30) to obtain an output audio signal with suppressed wind noise.

2. The method according to claim 1, further comprising: A wind noise indicator is determined (S2a, S2b, S2c) for each segment (101, 102, 103, 101', 102', 103') of the input audio signal (100, 100') using a wind noise detector (50), the wind noise indicator indicating at least one of a probability and an amplitude of wind noise in each segment (101, 102, 103, 101', 102', 103').

3. The method according to claim 2, further comprising: A wind noise state is determined (S3) based on the wind noise indicator. The method of claim 3 , wherein a filter coefficient of the high pass filter is based on the wind noise status.

5. The method according to claim 3 or 4, wherein each of the input audio signal (100, 100'), the wind noise reduced audio signal and the noise reduced audio signal comprises two audio channels, the method further comprising: Providing the wind noise state to a gain guidance module (11); If the wind noise conditions of the two channels exceed a first threshold level or if the difference between the wind noise conditions of the two channels is below a second threshold level: determining a common gain set based on a predicted gain set for at least one of the two channels; and Applying (S5b) the common gain set to both channels; otherwise: Each gain set is applied (S5b) to the corresponding channel.

6. The method according to claim 5, wherein: The common gain set is an average value, a maximum gain or a minimum gain of the gain set.

7. The method according to any one of claims 3 to 6, wherein determining (S2a, S2b, S2c) the wind noise state comprises: The wind noise indicator is provided to a state machine (61) having at least the following two states: No wind noise state NWN, and Wind noise maintains status WNH, wherein the state machine transitions to the WNH state in response to detecting a first number of subsequent segments (101, 102, 103, 101', 102', 103') associated with a wind noise indicator exceeding a high threshold, and outputs a high wind noise state at least until a next state change, and wherein the state machine transitions to the NWN state in response to detecting a second number of subsequent segments (101, 102, 103, 101', 102', 103') associated with wind noise indicators below a first low threshold and outputs a low wind noise state at least until a next state change.

8. The method according to claim 7, wherein the state machine (61) has four states: the NWN state, the WNH state, Wind noise attack state WNA, and Wind noise release state WNR, in, the state machine transitioning from the NWN state to the WNA state in response to detecting a segment (101, 102, 103, 101', 102', 103') having a wind noise indicator exceeding the high threshold, and outputting the low wind noise state at least until a next state change, wherein the state machine transitions from the WNA state to the WNH state in response to detecting the first number of subsequent segments (101, 102, 103, 101', 102', 103') associated with a wind noise indicator exceeding a high threshold, and outputs a high wind noise state at least until a next state change, wherein the state machine transitions from the WNH state to the WNR state in response to detecting a segment (101, 102, 103, 101', 102', 103') having a wind noise indicator below the first low threshold and outputs the high wind noise state at least until the next state change, Wherein, the state machine transitions from the WNR state to the NWN state in response to detecting that the second number of subsequent segments (101, 102, 103, 101', 102', 103') is below the first low threshold, and outputs the low wind noise state at least until the next state change.

9. The method according to claim 8, wherein: The state machine (61) transitions from the WNR state to the WNA state in response to detecting a segment (101, 102, 103, 101', 102', 103') associated with a wind noise indicator exceeding the high threshold and outputs the high wind noise state at least until the next state change.

10. The method according to claim 8 or 9, wherein detecting the first number of subsequent segments (101, 102, 103, 101', 102', 103') associated with the wind noise indicator exceeding the high threshold comprises: Counting the number of segments (101, 102, 103, 101', 102', 103') exceeding the high threshold with an attack counter 501; and resetting said attack counter 501 when a segment (101, 102, 103, 101', 102', 103') is associated with a wind noise indicator below a second lower threshold; The second low threshold is lower than the first low threshold.

11. The method according to any one of claims 3 to 10, wherein determining (S2a, S2b, S2c) the wind noise state comprises: The wind noise indicator is smoothed (S3) over at least two segments, wherein the wind noise state is based on the smoothed wind noise indicator.

12. A method according to any one of the preceding claims, wherein: Each gain set includes a plurality of gains associated with a corresponding frequency band in the audio channel.

13. The method according to any one of claims 1 to 12, wherein: The mixing of the wind noise reduced audio signal and the noise reduction audio signal is performed with a fixed mixing ratio (S6).

14. The method according to any one of claims 3 to 12, further comprising: For each segment (101, 102, 103, 101', 102', 103') in the input audio signal (100, 100'), a mixing ratio of the mixer (30) is controlled based on the wind noise state.

15. The method according to any one of claims 2 to 14, wherein the input audio signal (100, 100'), the wind noise reduced audio signal and the noise reduced audio signal comprise two audio channels, and in, The wind noise indicator is determined based on the monophonic characteristics of each individual audio channel and the differential characteristics associated with the two audio channels (S2b).

16. The method according to any one of claims 2 to 14, wherein the input audio signal (100, 100'), the wind noise reduced audio signal and the noise reduced audio signal comprise two audio channels, the method further comprising: determining (S2a) with a dual-mono detector (40) whether a corresponding segment of each of the two audio channels is a similar dual-mono segment or a different non-dual-mono segment based on a spectral power distribution of the segments in at least one frequency band; If the segment is a non-dual mono segment, determining (S2b) the wind noise indicator based on the mono feature and a differential feature associated with the two segments; Otherwise, the wind noise indicator is determined (S2c) based only on the monophonic features of each individual segment.

17. The method according to claim 16, wherein: The step of determining whether the two audio segments are similar dual-mono segments or different non-dual-mono segments comprises: determining a first sum as the sum of absolute values ​​of spectral differences between the two audio segments of the input audio signal (100, 100'); determining a second sum as the sum of the total spectral energy of each segment; calculating a ratio between the first sum and the second sum; determining that the segments are similar dual-mono segments if the ratio is below a predetermined ratio threshold and the second sum exceeds a predetermined sum threshold; and Otherwise, the segment is determined to be a different non-dual-mono segment.

18. The method according to any one of claims 2 to 17, further comprising: forming a non-real-time global wind noise indicator by aggregating the wind noise indicators of a plurality of segments (101, 102, 103, 101', 102', 103') of the input audio signal (100, 100'); Wherein, the filter coefficient of the high-pass filter is based on the global wind noise metric.

19. The method according to any one of claims 2 to 18, further comprising: forming a non-real-time global wind noise indicator by aggregating the wind noise indicators of a plurality of segments (101, 102, 103, 101', 102', 103') of the input audio signal (100, 100'); and For each segment (101, 102, 103, 101', 102', 103') of the input audio signal (100, 100'), a mixing ratio of the mixer is controlled based on the global wind noise metric.

20. A wind noise suppression system comprising a processor and a memory coupled to the processor, wherein the processor is adapted to perform the method steps according to any one of claims 1-19.

21. A computer program product comprising instructions which, when said program is executed by a computer, cause said computer to perform the method according to any one of claims 1 to 20.

22. A computer-readable storage medium storing the computer program according to claim 21.